Developer Investment in Agentic Platforms: Capability Spillovers, Hold-Up, and Portability

Abstract

Platforms increasingly host, distribute, and orchestrate third-party AI agents. A developer's investment in workflows, tool connections, testing, task records, and authorized memory creates a private capability stock and can also make complementary agents productive. The resulting stock is generated within a platform relationship, becomes platform specific, and is vulnerable to appropriation after investment. We model an origin platform that cannot commit to complete future fee and routing terms, a destination platform that commits to a verifiable recognition rule, and developers that invest before the origin resets terms. Portability protects investment through an exit option while preserving the origin's operating discretion. We characterize closed, partial, and full portability and show that the investment result survives incomplete ex post extraction. Verification has two opposing effects: it reduces transfer harm but can strengthen destination recognition and developer outside rent. It expands optimal portability only when safety and induced-investment gains exceed the additional rent and competitive leakage. Conditional on a target protected return, we compare portability with fee commitment and characterize the least-cost instrument mix. Finally, the origin is too closed when private leakage is mainly redistributive, but can be too open when duplication, fragmentation, or security costs are sufficiently real. The analysis turns agent interoperability into a conditional governance choice over relational capability rather than a presumption that openness is always efficient.